Niladri Das commited on
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -4,11 +4,122 @@ datasets:
|
|
| 4 |
language: en
|
| 5 |
library_name: custom
|
| 6 |
license: mit
|
|
|
|
| 7 |
tags:
|
| 8 |
- reinforcement-learning
|
| 9 |
- cartpole
|
| 10 |
- cma-es
|
| 11 |
- gymnasium
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
---
|
| 13 |
|
| 14 |
# Model Card for Model ID
|
|
@@ -25,10 +136,10 @@ tags:
|
|
| 25 |
|
| 26 |
|
| 27 |
|
| 28 |
-
- **Developed by:**
|
| 29 |
- **Funded by [optional]:** [More Information Needed]
|
| 30 |
- **Shared by [optional]:** [More Information Needed]
|
| 31 |
-
- **Model type:**
|
| 32 |
- **Language(s) (NLP):** en
|
| 33 |
- **License:** mit
|
| 34 |
- **Finetuned from model [optional]:** [More Information Needed]
|
|
|
|
| 4 |
language: en
|
| 5 |
library_name: custom
|
| 6 |
license: mit
|
| 7 |
+
model_name: CartPole-v1 CMA-ES Solution
|
| 8 |
tags:
|
| 9 |
- reinforcement-learning
|
| 10 |
- cartpole
|
| 11 |
- cma-es
|
| 12 |
- gymnasium
|
| 13 |
+
developers: Niladri Das
|
| 14 |
+
model_type: Linear Policy
|
| 15 |
+
evaluation: '
|
| 16 |
+
|
| 17 |
+
### Performance Metrics
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
The model achieves optimal performance:
|
| 21 |
+
|
| 22 |
+
- Perfect score (500.0) reached in 3 generations
|
| 23 |
+
|
| 24 |
+
- Consistent 500/500 steps across all test episodes
|
| 25 |
+
|
| 26 |
+
- Zero variance in performance during testing
|
| 27 |
+
|
| 28 |
+
- Population mean fitness improved from 41.59 to 484.64
|
| 29 |
+
|
| 30 |
+
'
|
| 31 |
+
model_description: "\nThe model implements a linear policy for the CartPole-v1 environment\
|
| 32 |
+
\ using CMA-ES optimization. \nIt maps the 4-dimensional state space to a 2-dimensional\
|
| 33 |
+
\ action space using 8 learned parameters (weights).\n"
|
| 34 |
+
training_data: '
|
| 35 |
+
|
| 36 |
+
The model was trained on the standard CartPole-v1 environment from Gymnasium.
|
| 37 |
+
|
| 38 |
+
'
|
| 39 |
+
model_summary: "\nThis model provides a solution to the CartPole-v1 environment using\
|
| 40 |
+
\ CMA-ES (Covariance Matrix Adaptation Evolution Strategy), \nachieving perfect\
|
| 41 |
+
\ performance with a simple linear policy.\n"
|
| 42 |
+
technical_specifications: '
|
| 43 |
+
|
| 44 |
+
### Model Architecture
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
Simple linear policy:
|
| 48 |
+
|
| 49 |
+
- Input: 4D state vector (cart position, cart velocity, pole angle, pole angular
|
| 50 |
+
velocity)
|
| 51 |
+
|
| 52 |
+
- Output: 2D action logits (move left/right)
|
| 53 |
+
|
| 54 |
+
- Architecture: Single linear transformation without bias
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
### Software Requirements
|
| 58 |
+
|
| 59 |
+
|
| 60 |
+
- Python 3.9+
|
| 61 |
+
|
| 62 |
+
- gymnasium
|
| 63 |
+
|
| 64 |
+
- numpy
|
| 65 |
+
|
| 66 |
+
- cma
|
| 67 |
+
|
| 68 |
+
'
|
| 69 |
+
limitations: '
|
| 70 |
+
|
| 71 |
+
- Solution specific to CartPole-v1 environment
|
| 72 |
+
|
| 73 |
+
- Linear policy may not generalize to more complex tasks
|
| 74 |
+
|
| 75 |
+
- Requires full state observation
|
| 76 |
+
|
| 77 |
+
'
|
| 78 |
+
intended_use: '
|
| 79 |
+
|
| 80 |
+
This model is designed for:
|
| 81 |
+
|
| 82 |
+
- Benchmarking CMA-ES against other RL algorithms
|
| 83 |
+
|
| 84 |
+
- Educational purposes in evolutionary strategies
|
| 85 |
+
|
| 86 |
+
- Baseline implementation for CartPole-v1
|
| 87 |
+
|
| 88 |
+
'
|
| 89 |
+
training_procedure: '
|
| 90 |
+
|
| 91 |
+
### Training Hyperparameters
|
| 92 |
+
|
| 93 |
+
|
| 94 |
+
- **Algorithm:** CMA-ES
|
| 95 |
+
|
| 96 |
+
- **Population size:** 16
|
| 97 |
+
|
| 98 |
+
- **Number of generations:** 15 (early convergence)
|
| 99 |
+
|
| 100 |
+
- **Initial step size:** 0.5
|
| 101 |
+
|
| 102 |
+
- **State dimensions:** 4
|
| 103 |
+
|
| 104 |
+
- **Action dimensions:** 2
|
| 105 |
+
|
| 106 |
+
- **Total parameters:** 8
|
| 107 |
+
|
| 108 |
+
|
| 109 |
+
### Training Results
|
| 110 |
+
|
| 111 |
+
|
| 112 |
+
- Convergence achieved in 3 generations
|
| 113 |
+
|
| 114 |
+
- Final population shows robust performance
|
| 115 |
+
|
| 116 |
+
- Training completed with perfect stability
|
| 117 |
+
|
| 118 |
+
'
|
| 119 |
+
citation_bibtex: "\n@misc{das2024cartpole,\n author = {Niladri Das},\n title = {CartPole-v1\
|
| 120 |
+
\ CMA-ES Solution},\n year = {2024},\n publisher = {Hugging Face},\n journal\
|
| 121 |
+
\ = {Hugging Face Model Hub},\n howpublished = {https://huggingface.co/bniladridas/cartpole-cmaes}\n\
|
| 122 |
+
}\n"
|
| 123 |
---
|
| 124 |
|
| 125 |
# Model Card for Model ID
|
|
|
|
| 136 |
|
| 137 |
|
| 138 |
|
| 139 |
+
- **Developed by:** Niladri Das
|
| 140 |
- **Funded by [optional]:** [More Information Needed]
|
| 141 |
- **Shared by [optional]:** [More Information Needed]
|
| 142 |
+
- **Model type:** Linear Policy
|
| 143 |
- **Language(s) (NLP):** en
|
| 144 |
- **License:** mit
|
| 145 |
- **Finetuned from model [optional]:** [More Information Needed]
|